The Agent Skills Directory

Data Exposure (LOW): The scripts/snapshot_git_state.sh tool captures repository metadata, including git remote -v and git reflog. These files are stored in a local .git-state-snapshots directory. This metadata can contain sensitive information such as API tokens in remote URLs or historical data in the reflog. This exposure is intrinsic to the skill's purpose of Git state recovery.
Indirect Prompt Injection (LOW): The scripts/guided_repair_plan.sh script parses local Git output (e.g., branch names, worktree paths). A malicious repository could craft metadata to influence the script's diagnostic logic.
Ingestion points: Reads worktree_list.txt and status.txt generated from local Git commands in scripts/snapshot_git_state.sh.
Boundary markers: None identified.
Capability inventory: The skill suggests Git commands and provides a playbook for manual .git/HEAD modification.
Sanitization: No sanitization of Git output is performed before parsing.
Command Execution (SAFE): The skill utilizes standard Git commands for its operations. All command execution is focused on repository management and is consistent with the skill's stated goal of repository recovery.

unfuck-my-git-state